What data does this category have
Core data for the sheet metal cutting and bending category comes primarily from CAD/CAM software, order BOMs, sheet metal inventory systems, laser cutters, bending machines, and MES systems. Daily generated documents include 3D part unfolding drawings, CNC cutting programs, bending process cards, nesting diagrams, first-piece inspection records, and process flow cards. Data updates trigger per order, daily production schedule, or each production line changeover. Common document fields include sheet thickness, bending compensation, cutting dimensions, and inventory remaining quantity. Units are typically millimeters, kilograms, or sheets.
What constraints do these characteristics impose on document parsing and chunking
Documents for this category are mostly engineering drawings or process files with structured parameters. Mixed-format input scenarios are common, such as PDFs exported from CAD, TXT files for CNC programs, and DOCX process cards. The multi-variety, small-batch business model leads to high document volume per batch and frequent updates. Bid scenarios require fast processing of multiple bid documents, placing high demands on parallel parsing capabilities. Documents contain many specialized engineering parameters. Chunking must retain parameter associations to avoid losing critical information like bending coefficients and cutting dimensions. For internal network deployment scenarios, parsing support for internal document sources such as Confluence is needed. Single-file parsing duration must also be controlled to match production line changeover response timelines.
How to set configurations
This explanation uses the configuration logic of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the per-paragraph parameter length of process cards and nesting diagrams in sheet metal documents, avoiding loss of association between bending compensation and cutting dimensions after chunking |
chunkSplitter | By paragraph + fixed length | Adapts to the natural paragraph structure of engineering documents, preserving the integrity of process steps and parameter descriptions |
customPdfParse | Enable CAD-exported PDF parsing mode | For PDFs exported from sheet metal unfolding drawings, retains dimension annotations and parameter text in vector graphics |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapts to the parsing duration of large nesting diagrams and multi-step process flow cards, preventing file parsing failures due to timeout |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports upload requirements for single large nesting diagrams or multi-page process collections, adapting to batch file processing scenarios for multi-variety, small-batch work |
| Chunk overlap length | 50–80 characters | Retains process association information between adjacent paragraphs, avoiding logical breaks in bending sequence after chunking |
Three common configuration mistakes
- Parsing results produce unformatted full text with no hierarchical structure. Cause: The engineering document mode of
customPdfParseis not enabled, and only general PDF parsing logic is used. This causes loss of title and parameter list formatting in process cards. - Internal Confluence pages cannot be parsed for content. Cause: Internal network access adaptation rules are not configured. Only public external web page crawling is supported, and custom internal source parsing is not enabled.
- Parent-child document association relationships cannot be retained after chunking. Cause: Parent-child chunk configuration is not enabled. Split document blocks cannot associate the binding relationship between the original nesting diagram and the corresponding process card.
How to confirm the configuration is correct
- Upload a bending process card DOCX file, and check if the parsed text retains the hierarchical format of titles, process lists, and parameter items.
- Upload an unfolding drawing PDF exported from CAD, and confirm that the parsed result contains key parameter text such as sheet thickness and bending compensation.
- Submit an internal Confluence page link, and verify that the page content can be normally pulled and chunked.
- Upload a single nesting diagram file that complies with the
UPLOAD_FILE_MAX_SIZElimit, and confirm that parsing does not trigger timeout or upload failure prompts.
The values listed above are common starting points and should be evaluated against samples from the actual deployment context.